It’s happening; AI has moved beyond the Hype Stage in Tauranga

At the beginning of the year, for our February Tauranga Software Test Professionals Meetup, TTC Global conducted a brief survey on the state of Software Quality Assurance in Tauranga. Amongst the questions asked was one on the likelihood of the use of AI in Software Quality Assurance Delivery. 100% of respondents expected this to be the case. In our June meetup, hosted by the great team at Zespri International, three local organisations showcased exactly how it was being done.

And it was very good.

First up was Tauranga City Council. They have been an early adopter of the use of AI in the delivery of software quality assurance. Their presentation back in February 2025 was memorable not just for the early insight into a production-level use of AI for Test Case Generation, but also for the AI-generated country and western song about the meetup agenda 😊. Rodney together with Daniel presented an update of their AI journey – now a focus on real time dashboards that enable project leaders to get insight on quality assurance metrics and predictions for their work programme. A key feature of their use of AI was the collaborative and rapid development of dashboard features using AI – what would take weeks if not months being delivered within half a day making the prototype to production process easily deliverable and configurable. Their presentation was a live demo of the dashboards that they have developed – some screen shots from that presentation are in their AI Presentation here.

Zespri International was our next presenter. Like Tauranga City Council, Zespri has been a regular contributor to the Tauranga Software Test Professionals Meetup. And a leading innovator. Their presentation continued this trend. Undergoing a major core systems upgrade, Zespri’s focus has been on delivery efficiency – both from a cost and a timeframe perspective yet with no impact on quality and ideally a quality uplift. AI in quality assurance does away with the constraints of the time / cost / quality triangle. You can get three out of three vectors now. Being an analytical person, Berny’s presentation also included evidenced RoI on their use of AI, often missed in real world AI deployments. With a 44% reduction in effort equating to a saving of 29 days, the net effect is that AI is delivering noticeable productivity improvements, enabling the same team to deliver more. Like Tauranga City Council, Zespri has also leveraged AI for the development and maintenance of KPI and metrics dashboards, improvising visibility of software testing status and software delivery times. Berny’s presentation is available here.

To round off the evening, we were stoked to hear from Charlotte of Atomic Labs. Having Charlotte present was a long time coming – and very much worth the wait. Charlotte has been working with Agentic AI for a while now – an earlier description of the environment that Charlotte had set up was very exciting:

Our agents build context from GitHub tickets, the codebase, and our test repositories (via GitHub MCP and Open Code). They plan, generate, run and iterate on tests, particularly for new features. We’re also starting to factor in runtime signals — high-traffic routes and error hotspots from logs — to move toward more risk-informed prioritisation rather than relying purely on structural coverage.

In terms of scope, the agents handle planning through to execution and refinement. That said, we keep humans firmly in the loop. Every commit is reviewed. There are checkpoints in the planning flow, and if an agent fails to resolve something after a few attempts, it exits and flags for human input. We deliberately added those break conditions to avoid runaway loops and token waste.

For orchestration, we’ve been experimenting with different patterns. One that’s working well uses “oh my opencode” as the planning layer, with subordinate agents handling execution tasks. Introducing clearer personas and boundaries significantly improved behaviour — earlier versions of the coding agents would sometimes introduce failing tests and attribute them to unrelated issues, which turned out to be a context transfer problem. Tightening orchestration helped reduce that.

We’ve also moved from the Playwright MCP to the Playwright Skills CLI due to token efficiency. One practical advantage over tools like Copilot is that we can refine context while the workflow is running, rather than waiting for completion.

Charlotte’s presentation, available here, not only included a demonstration of the environment described above, it provided an overview of the framework of the Agentic AI implementation that Charlotte had developed. Skills encapsulated into agents – complete with guardrails means that the skills can be maintained over time separate to the Agentic AI workflows. And the skills could be called into the Agentic AI workflows in any order / sequence and frequency meaning an ability to create a range of Agentic AI processes quickly, easily and safely. More so when the skills were developed in a way that they checked in with the human that was overseeing the workflow to ensure that drift was monitored, measured, managed and mitigated where needed. Eye opening stuff when seen in a production / actually used form.

All in all – three very impressive presentations. And lots of post talk discussion and idea sharing.

The Tauranga meetup community continues to innovate and collaborate. After seven years of working with this community, right from our first meetup (ever) in 2019, it was only fitting that this one in June 2026 also closed out my involvement with the community, given my resignation from TTC Global.

It has been a blast!!